The account is locked. Does the AI still remember what they told it?

#What gets deleted and what gets handed over is decided in advance

A sales rep has left your company. You locked the account, but everything that person told the AI over several months—client circumstances, their own responsibilities—is still there, along with their conversations with it. An AI that learns more about your people and data the more it is used is convenient, but it gives your security team more to work through. Where does that memory live, who can see it, and who deletes it when an employee leaves or a contract ends?

The answer depends on whose memory it is. What the person shared with the AI is deleted according to the offboarding policy an administrator sets in advance; what they built for the team passes to the organization. OwlAI remembers in three ways: personal memory about one person, table notes that attach to a table and are shared by the team, and search over past conversations. All three are stored per organization in the installation's own database, and each has a fixed answer to who can see it and when it is deleted.

#Memory lives in the database of your installation

Memory is stored in the database of the environment where OwlAI is installed. On-premises, that means your own servers; with SaaS, the cloud environment that OwlAI operates. There is no separate external store for memory. Building the search index for memory does pass through the AI model configured for that installation, so an installation that uses only internal models, such as one on an air-gapped network, keeps that step in-house too.

All memory is stored separately for each organization. If one person belongs to several organizations, their personal memory is kept separately for each, and memory built up in one organization does not leak into conversations in another.

#Only you see your personal memory; only those with table access see its notes

Personal memory is used only in that person's own conversations, and they can view and edit it on the settings screen. Every change—what was added, what was removed—goes into the audit log, so if the AI remembers something odd, you can trace when and how it got there. The log records whether the AI or the person made the change and which lines changed. If the person clears their memory entirely, the log keeps only the time and the number of lines, not the content. This log is kept only in your installation's database and does not appear in the activity log of the admin console.

Table notes are included only in answers for people who have permission to use that table. People who cannot access the table do not see its notes either. Administrators can review and delete notes on the database screen.

Past-conversation search looks only at the person's own conversations. Colleagues' and administrators' conversations are not searched, and the questions and answers in conversations do not appear in the administrator's activity log either.

#A deleted conversation does not come back through search

When a user deletes a conversation, it disappears from the list and from search right away, and under the default settings it is permanently removed from storage 30 days later. Conversations with no activity for a year are also deleted under the default settings. Deleting a conversation also deletes the content prepared for searching it, so a deleted conversation does not resurface in search results.

When an organization ends its service, its personal memory, table notes, and conversation records are all deleted once the retention period is over. Records that must be kept by law, such as audit logs, remain for the required retention period.

#Team work is handed over; conversations with the AI are deleted

What a departing person built for the team passes to the organization. Playbooks they created and documents they uploaded become the organization's property, and the next person picks them up as they are. Table notes are the team's knowledge about its data, not one person's memory, so they stay after the author leaves; only the author label becomes anonymous. A senior colleague's "watch the currency unit" note keeps protecting the next person's answers after that colleague is gone.

What they shared with the AI is theirs. Personal memory, conversations, and the index for past-conversation search are not opened even to administrators, and they are deleted according to the offboarding policy the administrator sets. The policy is one of two options: delete immediately, or delete after a set retention period. A company that keeps no personal records as a rule would pick immediate deletion; one that wants time to tidy up would pick deletion after retention.

Because administrators do not see conversations, employees can ask the AI without wondering who might read their questions later. The flip side is that the company does not inherit conversations, so the business knowledge it needs to keep has to be left in places the team shares: playbooks, documents, and table notes.

When offboarding is complete, the audit log records what was deleted and what was handed over. Deleting an account also deletes the person's personal memory and the records behind conversation search. A disabled account picks up where it left off when it is turned back on; the offboarding policy applies only when the person is actually removed from the organization.

#What to decide before rollout

An administrator has only a few things to settle at rollout:

• Who reviews table notes on a regular schedule
• Which offboarding policy applies when members leave (delete immediately, or delete after retention)
• How employees learn that they can check their own memory under "Memory" in settings